Fea/support cmake cuda architectures rapids value (#327)

CMake 3.23+ now offers support for `all` and `all-major` as special keywords to `CMAKE_CUDA_ARCHITECTURES`. This means that rapids-cmake overload on `ALL` now creates confusion and breaks expectations ( trxcllnt/rapids-compose#84 https://gitlab.kitware.com/cmake/cmake/-/issues/23739, #104 ) 

This PR introduces the new magic keyword of `RAPIDS` which behaves the same as `ALL`, and will become the new apporved approach for projects that want to build for RAPIDS supported platforms.

Fixes #314


More details on the plans of deprecation for `ALL` can be found at: #318

Authors:
  - Robert Maynard (https://github.com/robertmaynard)

Approvers:
  - Bradley Dice (https://github.com/bdice)

URL: #327

docs/command/supported_cuda_architectures_values.txt

@@ -0,0 +1,7 @@
+
+``NATIVE`` or ``""``:
+  When passed as the value for :cmake:variable:`CMAKE_CUDA_ARCHITECTURES <cmake:variable:CMAKE_CUDA_ARCHITECTURES>` or :cmake:envvar:`ENV{CUDAARCHS} <cmake:envvar:CUDAARCHS>`
+  will compile for all GPU architectures present on the current machine.
+
+``RAPIDS``, ``ALL``, or no value in :cmake:variable:`CMAKE_CUDA_ARCHITECTURES <cmake:variable:CMAKE_CUDA_ARCHITECTURES>` and :cmake:envvar:`ENV{CUDAARCHS} <cmake:envvar:CUDAARCHS>`:
+  When passed as the value for :cmake:variable:`CMAKE_CUDA_ARCHITECTURES <cmake:variable:CMAKE_CUDA_ARCHITECTURES>` or :cmake:envvar:`ENV{CUDAARCHS} <cmake:envvar:CUDAARCHS>` will compile for all supported RAPIDS GPU architectures.

rapids-cmake/cuda/detail/invoke_set_all_architectures.cmake

@@ -23,4 +23,4 @@ endif()
 # Used by rapids_cuda_init_architectures to allow the `project()` call to invoke the
 # rapids_cuda_set_architectures function after compiler detection
 #
-rapids_cuda_set_architectures(ALL)
+rapids_cuda_set_architectures(RAPIDS)

rapids-cmake/cuda/init_architectures.cmake

@@ -29,9 +29,9 @@ to include support for `ALL` and `NATIVE` to make CUDA architecture compilation
     rapids_cuda_init_architectures(<project_name>)
 
 Used before enabling the CUDA language either via :cmake:command:`project() <cmake:command:project>` to establish the
-CUDA architectures to be compiled for. Parses the :cmake:envvar:`CUDAARCHS <cmake:envvar:CUDAARCHS>`, and
+CUDA architectures to be compiled for. Parses the :cmake:envvar:`ENV{CUDAARCHS} <cmake:envvar:CUDAARCHS>`, and
 :cmake:variable:`CMAKE_CUDA_ARCHITECTURES <cmake:variable:CMAKE_CUDA_ARCHITECTURES>` for special values
-`ALL`, `NATIVE` and `""`.
+`ALL`, `RAPIDS`, `NATIVE` and `""`.
 
 .. note::
   Required to be called before the first :cmake:command:`project() <cmake:command:project>` call.
@@ -43,15 +43,7 @@ CUDA architectures to be compiled for. Parses the :cmake:envvar:`CUDAARCHS <cmak
 ``project_name``
   Name of the project in the subsequent :cmake:command:`project() <cmake:command:project>` call.
 
-``NATIVE`` or ``""``:
-  When passed as the value for :cmake:variable:`CMAKE_CUDA_ARCHITECTURES <cmake:variable:CMAKE_CUDA_ARCHITECTURES>`
-  will compile for all GPU architectures present on the current machine.
-
-``ALL`` or no :cmake:variable:`CMAKE_CUDA_ARCHITECTURES <cmake:variable:CMAKE_CUDA_ARCHITECTURES>` and
-  :cmake:envvar:`ENV{CUDAARCHS} <cmake:envvar:CUDAARCHS>`:
-  When passed as the value for :cmake:variable:`CMAKE_CUDA_ARCHITECTURES <cmake:variable:CMAKE_CUDA_ARCHITECTURES>`
-  will compile for all supported RAPIDS GPU architectures.
-
+.. include:: supported_cuda_architectures_values.txt
 
 Example on how to properly use :cmake:command:`rapids_cuda_init_architectures`:
 
@@ -79,21 +71,23 @@ function(rapids_cuda_init_architectures project_name)
   # If `CMAKE_CUDA_ARCHITECTURES` is not defined, build for all supported architectures. If
   # `CMAKE_CUDA_ARCHITECTURES` is set to an empty string (""), build for only the current
   # architecture. If `CMAKE_CUDA_ARCHITECTURES` is specified by the user, use user setting.
-  if(DEFINED ENV{CUDAARCHS} AND "$ENV{CUDAARCHS}" STREQUAL "ALL")
-    set(cuda_arch_mode "ALL")
+  if(DEFINED ENV{CUDAARCHS} AND ("$ENV{CUDAARCHS}" STREQUAL "RAPIDS" OR "$ENV{CUDAARCHS}" STREQUAL
+                                                                        "ALL"))
+    set(cuda_arch_mode "RAPIDS")
   elseif(DEFINED ENV{CUDAARCHS} AND "$ENV{CUDAARCHS}" STREQUAL "NATIVE")
     set(cuda_arch_mode "NATIVE")
-  elseif(CMAKE_CUDA_ARCHITECTURES STREQUAL "ALL")
-    set(cuda_arch_mode "ALL")
+  elseif(CMAKE_CUDA_ARCHITECTURES STREQUAL "RAPIDS" OR CMAKE_CUDA_ARCHITECTURES STREQUAL "ALL")
+    set(cuda_arch_mode "RAPIDS")
   elseif(CMAKE_CUDA_ARCHITECTURES STREQUAL "" OR CMAKE_CUDA_ARCHITECTURES STREQUAL "NATIVE")
     set(cuda_arch_mode "NATIVE")
   elseif(NOT (DEFINED ENV{CUDAARCHS} OR DEFINED CMAKE_CUDA_ARCHITECTURES))
-    set(cuda_arch_mode "ALL")
+    set(cuda_arch_mode "RAPIDS")
   endif()
 
-  # This needs to be run before enabling the CUDA language since RAPIDS supports the magic string of
-  # "ALL"
-  if(cuda_arch_mode STREQUAL "ALL")
+  # This needs to be run before enabling the CUDA language since RAPIDS supports magic values like
+  # `RAPIDS`, `ALL`, and `NATIVE` which if propagated cause CMake to fail to determine the CUDA
+  # compiler
+  if(cuda_arch_mode STREQUAL "RAPIDS")
     set(CMAKE_CUDA_ARCHITECTURES OFF PARENT_SCOPE)
     set(load_file "${CMAKE_CURRENT_FUNCTION_LIST_DIR}/detail/invoke_set_all_architectures.cmake")
   elseif(cuda_arch_mode STREQUAL "NATIVE")

rapids-cmake/cuda/set_architectures.cmake

@@ -39,13 +39,7 @@ directly.
 .. note::
   This is automatically called by :cmake:command:`rapids_cuda_init_architectures`
 
-``NATIVE``:
-  When passed NATIVE as the first parameter will compile for all
-  GPU architectures present on the current machine. Requires that
-  the CUDA language be enabled for the current CMake project.
-
-``ALL``:
-  When passed ALL will compile for all supported RAPIDS GPU architectures
+.. include:: supported_cuda_architectures_values.txt
 
 Result Variables
 ^^^^^^^^^^^^^^^^
@@ -66,7 +60,7 @@ function(rapids_cuda_set_architectures mode)
     list(REMOVE_ITEM supported_archs "90")
   endif()
 
-  if(${mode} STREQUAL "ALL")
+  if(${mode} STREQUAL "RAPIDS" OR ${mode} STREQUAL "ALL")
 
     # CMake architecture list entry of "80" means to build compute and sm. What we want is for the
     # newest arch only to build that way while the rest built only for sm.

testing/cuda/CMakeLists.txt

@@ -27,6 +27,8 @@ add_cmake_config_test( init_arch-existing-project-flags.cmake )
 add_cmake_config_test( init_arch-native.cmake )
 add_cmake_config_test( init_arch-native-via-empty-str )
 add_cmake_config_test( init_arch-native-via-env.cmake )
+add_cmake_config_test( init_arch-rapids.cmake )
+add_cmake_config_test( init_arch-rapids-via-env.cmake )
 add_cmake_config_test( init_arch-user.cmake )
 add_cmake_config_test( init_arch-user-via-env.cmake )
 
@@ -37,3 +39,4 @@ add_cmake_config_test( set_arch-all.cmake )
 add_cmake_config_test( set_arch-existing.cmake )
 add_cmake_config_test( set_arch-invalid-mode.cmake )
 add_cmake_config_test( set_arch-native.cmake )
+add_cmake_config_test( set_arch-rapids.cmake )

testing/cuda/init_arch-all-via-env.cmake

@@ -29,4 +29,4 @@ if(CMAKE_CUDA_ARCHITECTURES STREQUAL "ALL")
 endif()
 
 
-include("${rapids-cmake-testing-dir}/cuda/validate-cuda-all.cmake")
+include("${rapids-cmake-testing-dir}/cuda/validate-cuda-rapids.cmake")

testing/cuda/init_arch-all-via-undef/CMakeLists.txt

@@ -30,4 +30,4 @@ if(CMAKE_CUDA_ARCHITECTURES STREQUAL "ALL")
   message(FATAL_ERROR "rapids_cuda_init_architectures didn't init CUDA_ARCHITECTURES")
 endif()
 
-include("${rapids-cmake-testing-dir}/cuda/validate-cuda-all.cmake")
+include("${rapids-cmake-testing-dir}/cuda/validate-cuda-rapids.cmake")

testing/cuda/init_arch-all.cmake

@@ -29,4 +29,4 @@ if(CMAKE_CUDA_ARCHITECTURES STREQUAL "ALL")
 endif()
 
 
-include("${rapids-cmake-testing-dir}/cuda/validate-cuda-all.cmake")
+include("${rapids-cmake-testing-dir}/cuda/validate-cuda-rapids.cmake")

testing/cuda/init_arch-rapids-via-env.cmake

@@ -0,0 +1,32 @@
+#=============================================================================
+# Copyright (c) 2022, NVIDIA CORPORATION.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#=============================================================================
+include(${rapids-cmake-dir}/cuda/init_architectures.cmake)
+
+
+set(ENV{CUDAARCHS} "RAPIDS")
+rapids_cuda_init_architectures(rapids-project)
+project(rapids-project LANGUAGES CUDA)
+
+if(NOT DEFINED CMAKE_CUDA_ARCHITECTURES)
+  message(FATAL_ERROR "CMAKE_CUDA_ARCHITECTURES should exist after calling rapids_cuda_init_architectures()")
+endif()
+
+if(CMAKE_CUDA_ARCHITECTURES STREQUAL "RAPIDS")
+  message(FATAL_ERROR "rapids_cuda_init_architectures didn't init CUDA_ARCHITECTURES")
+endif()
+
+
+include("${rapids-cmake-testing-dir}/cuda/validate-cuda-rapids.cmake")

testing/cuda/init_arch-rapids.cmake

@@ -0,0 +1,32 @@
+#=============================================================================
+# Copyright (c) 2021, NVIDIA CORPORATION.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#=============================================================================
+include(${rapids-cmake-dir}/cuda/init_architectures.cmake)
+
+
+set(CMAKE_CUDA_ARCHITECTURES "RAPIDS")
+rapids_cuda_init_architectures(rapids-project)
+project(rapids-project LANGUAGES CUDA)
+
+if(NOT DEFINED CMAKE_CUDA_ARCHITECTURES)
+  message(FATAL_ERROR "CMAKE_CUDA_ARCHITECTURES should exist after calling rapids_cuda_init_architectures()")
+endif()
+
+if(CMAKE_CUDA_ARCHITECTURES STREQUAL "RAPIDS")
+  message(FATAL_ERROR "rapids_cuda_init_architectures didn't init CUDA_ARCHITECTURES")
+endif()
+
+
+include("${rapids-cmake-testing-dir}/cuda/validate-cuda-rapids.cmake")

testing/cuda/set_arch-all.cmake

@@ -21,4 +21,4 @@ if(NOT DEFINED CMAKE_CUDA_ARCHITECTURES)
   message(FATAL_ERROR "CMAKE_CUDA_ARCHITECTURES should exist after calling rapids_cuda_set_architectures()")
 endif()
 
-include("${rapids-cmake-testing-dir}/cuda/validate-cuda-all.cmake")
+include("${rapids-cmake-testing-dir}/cuda/validate-cuda-rapids.cmake")

testing/cuda/set_arch-rapids.cmake

@@ -0,0 +1,24 @@
+#=============================================================================
+# Copyright (c) 2021, NVIDIA CORPORATION.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#=============================================================================
+include(${rapids-cmake-dir}/cuda/set_architectures.cmake)
+
+rapids_cuda_set_architectures(RAPIDS)
+
+if(NOT DEFINED CMAKE_CUDA_ARCHITECTURES)
+  message(FATAL_ERROR "CMAKE_CUDA_ARCHITECTURES should exist after calling rapids_cuda_set_architectures()")
+endif()
+
+include("${rapids-cmake-testing-dir}/cuda/validate-cuda-rapids.cmake")